← ClaudeAtlas

cinematic-reellisted

Turn one raw landscape/cinematic clip (DJI Osmo Pocket, drone, phone) into a polished vertical 9:16 cinematic reel — scale/crop, color grade, beat-snapped chapter/title cards, optional ElevenLabs VO + ducked BGM — in a SINGLE ffmpeg pass (NVENC with libx264 auto-fallback), then optionally auto-schedule it across all socials via GHL MCP with a different human caption + hashtag set per platform. Distilled from a SkynetLabs sunset/landscape reel (2026-05-29). Trigger when user says "cinematic reel", "/cinematic-reel", "make a reel from this clip", "sunset/landscape reel", "cut this DJI footage", drops a single scenic .mp4 and wants a vertical cinematic cut, or wants a reel rendered AND scheduled to socials.
waseemnasir2k26/skynetlabs-all-claude-code · ★ 0 · AI & Automation · score 62
Install: claude install-skill waseemnasir2k26/skynetlabs-all-claude-code
# Cinematic Reel — scenic clip → vertical cut → all-socials schedule End-to-end for a SINGLE scenic source clip (sunset, landscape, b-roll, drone, Osmo Pocket). Output: vertical 9:16 cinematic cut with title/chapter cards + optional VO/BGM, then optional multi-platform GHL scheduling. **Single ffmpeg pass.** ## When to use - One scenic/cinematic source clip → want a vertical reel with grade + text cards. - Optionally: schedule it to FB/IG/TikTok/YT/LinkedIn/Pinterest at staggered times with per-platform copy. Not for: talking-head/script reels (→ `reel-studio`), landscape screencast promos (→ `demo-promo`), static social cards (→ `social-stack`). ## Pipeline ### 1. Probe + verify the arc (DO NOT skip) `ffprobe` the source (res/fps/duration). Then extract frames across the timeline and **eyeball them** — card text must match what's actually on screen. Numeric/agent color analysis lies (auto-exposure, artificial lights). See `references/pipeline-notes.md`. ``` ffmpeg -ss <t> -i SRC -frames:v 1 -q:v 3 frame_<t>.jpg ``` ### 2. Write the config + render (one pass) Build a `reel.config.json` (schema in `scripts/build_cut.py` header). Snap card `t_in` to the BGM beat grid; card at VO handoff, one before the ~50% drop-off, ~50-60s cadence. Then: ``` python -u <skill>/scripts/build_cut.py reel.config.json ``` `encoder:"auto"` tries h264_nvenc and falls back to libx264 if the GPU driver is too old. Audio comes out 48kHz stereo, loudnorm ~-14. ### 3. QA the render Probe st